When magnetic resonance imaging (MRI) is used as a modality for guiding a minimally invasive procedures, materials used to fabricate catheters or biopsy needles have to be chosen such that their magnetic susceptibility does not distort the acquired MR images. Such susceptibility artefacts can make MRI unusable or harm its use as a guiding modality in minimally invasive interventions such as liver or breast biopsies. It can affect the image quality if the area of interest is close to the position of the passive instrument , .
To not significantly affect the quality of the diagnostic information of the MR images, testing of susceptibility artifacts arising from passive instruments is important before introducing them into the MRI environment , . Therefore, the precise measurement of the artefact size is important during the development stage of medical tools to be used in MRI. The quantification of artefacts caused by such devices is regulated by the ASTM F2119 standard . The ASTM defines a quantitative but manual method to assess the artefact, where the definition of the artefact is clearly defined but gives only an approximative method for its quantification . Hence, it can be assumed that observed artefact dimensions are user dependent despite a standard exists for reducing subjectivity.
Some authors have proposed different methods to assess artifacts using mainly software measurement tools , , while other works proposed to automatize the standard measurement objective and reproducible measurement that comply with this standard , , .
This work investigates whether an automatic algorithm can be used as an alternative for measuring susceptibility artefacts of passive instruments used in MRI. The algorithm is based on the computation of a quantitative indicator related to the amount of distortion produced by an instrument on a three-dimensional (3D) surface obtained from the 2D MR image.
For the evaluation of the proposed algorithm, six MRI compatible biopsy needles were used for artefact assessment. The results show that an indicator can be computed automatically from all the slices of a given MRI sequence and that a value computed from the indicator is correlated with the average of the manual measurements.
2.1 Data acquisition and manual measurements setup
Six different MR compatible biopsy needles with outer diameters of 1 mm to 1.3 mm (ITP GmbH, Somatex GmbH, Germany) were imaged in a 3T MRI scanner (Magnetom Skyra, Siemens) using a FLASH sequence (see Table 1). A Plexiglas phantom filled with a solution of water and copper sulphate was used for the experiments .
Artefacts were manually analysed by three MRI or image analysis experts. For each needle configuration, the participants had to select one slice for measuring the artefact size. Each artefact was measured twice by each participant (with a pause of one day between the measurements) following  using a DICOM viewer (RadiAnt DICOM). In summary, every participant made six artefact measurements in each of the two measurement series and therefore six measurements are available for each needle.
2.2 Image distortion indicator computation
The standard 2D MR image depicted in Figure 1A shows the artefact of a biopsy needle. The artefact can be represented as a 3D surface version of the MR image by representing the values of each pixel (being located in the x-y plane) along the z-axis (see Figure 1B). In this 3D surface version the susceptibility artefact can be viewed as a distortion of the 3D surface image background. This distortion appears as a downward deflection of the background. It seems that the artefact observed in a 3D surface point of view has a more complicated nature than the traditional 2D image, therefore the main idea is to quantify this deflection of the background using a quantitative indicator of background distortion. For that we propose to calculate an indicator related to the volume resulting from the background deflection produced by the artefact in the 3D surface image.
Before doing this, it is important to observe in Figure 1B that the surface baseline is located above zero and that it has non-homogeneous characteristics with low frequency trends. This last feature can be explained by the location of the Plexiglas phantom with respect to the RF coil position and its inhomogeneous field distribution, which causes a variation of the signal intensity.
Therefore it is required first to attenuate the low frequency trend of the background surface and to remove the surface offset. Thus, the deflection volume, where no artefact is present, will tend to be zero, while the volume where the artefact is present will tend to be negative: the more negative the indicator value is, the more pronounced is the artefact level.
To remove the low frequency trend and the offset, a preprocessing step is performed over each horizontal signal line of the surface image and then the 3D surface is reconstructed using the preprocessed signal lines. In a first step the horizontal signal line passes through a moving average (MA) smoother. Then the deflection waveform (if it is present) is removed from the signal line and an hermite interpolation is used to fill the removed signal segment. A Kalman smoother  is then used for the low frequency trend extraction, which is finally subtracted to the MA smoothed signal. In Figure 1C the obtained 3D surface image is shown after preprocessing of every signal line and surface reconstruction.
The resulting preprocessed 3D surface of Figure 1C is only the result for one MRI slice. The main algorithm processes every signal line of every slice for calculating a final distortion indicator. A summary of this algorithm is presented in the following:
For each slice the preprocessing algorithm described above is applied.
For each line of the preprocessed 3D surface image of each image slice, an indicator of artefact level is computed using a integration operation over the preprocessed signal line.
All the values of the indicator of every signal line and slice are stored in the same vector.
The elements of the vector are sorted in descending order.
Figure 2 shows an example of the distortion indicator per slice when the algorithm was applied to 13 slices of a needle MRI image, where it is possible to observe how the indicator changes according to the visibility of the artefact. At the right of this figure, the descending order sorted vector is shown, which correspond to the final distortion indicator. In this way the artefact assessment results in a monotonically decreasing curve (MDC) indicator, which goes from zero values representing signal lines, where no artefact is present, passing through middle values and arriving to lower negative values representing the maximal artefact levels assessed in different signal lines. These lower values are the ones which are significant for assessing the artefact created by the tested instrument.
First of all in Figure 3 the MDC indicator is shown for every tested biopsy needle (named N1, …, N6 in the sequel). Since the MDC curve is the result of a descending sorted signal, only the last part of this indicator is significant (the maximal value of artefact assessment is in this section of the indicator). That is why only the last 150 values of the MDC signal are shown. Despite that, due to the nature of the indicator the MDC, it has no distance unit but it shows clearly the different artefact sizes produced by each needle.
For comparison between manual and automatic artefact assessment, the measured manual artefact were evaluated for each needle using the average and standard deviation of all the manual measurements performed per needle. This is shown at the left of Figure 4, where the average and standard deviation of the six measures per needle (2 per each participant) result in different artefacts levels. It is also possible to observe significant variability between observers with sometimes more than 1 mm of measurements differences between them. At the right of the same figure the automatic measurements are shown, where the displayed single values per needle were calculated as the average of the last 100 values of the obtained MDC for each needle. With the exception of the differences in artefact between needles N1 and N6 the curves have an acceptable correlation.
The variability between participants shown in the left of Figure 4 is not only caused by different interpretations on how to measure an artefact using the ASTM standard, but also because the observers selected different slices for performing their measurements. This is revealed by Table 2, where the selected slices for each needle by each participant are shown. It is evident that not always the same slice is selected by the participants and that moreover sometimes not the same slice is selected by the same observer in two different measurements of the same artefact. The last row of the table shows the slice where the automatic algorithm finds the largest artefact. It can be noticed that this maximal artefact measurement is in almost all cases located in the slice where there is a manual selection agreement of at least 50% of all the measurements performed in a needle.
The presented work had the objective of comparing artefact assessment performed by manual measurements with an automatic algorithm where the algorithms measurement principle is based on a different point of view than the proposed standards. Assessing the artefact through the computation of a 3D surface distortion indicator shows a correlation with the manually obtained results.
One question to answer in the future is if the 3D representation of the artefact can reveal more complex structures than the 2D-images can and if this can improve the accuracy in artefact assessment. Also, further work is needed to use the indicator for quantitative information about the artefact in terms of an accepted physical unity.
Research funding: This research was funded by the German BMBF (grant 03IPT7100X). Conflict of interest: Authors state no conflict of interest. Material and methods: Informed consent: Informed consent is not applicable. Ethical approval: The conducted research is not related to either human or animal use.
Kaiser M, Detert M, Rube MA, El-Tahir A, Elle O J, Melzer A, et al. Resonant marker design and fabrication techniques for device visualization during interventional magnetic resonance imaging. Biomed Technik. 2015;60:89–103. Google Scholar
Patil S, Bieri O, Jhooti P, Scheffler K. Automatic slice positioning (ASP) for passive real-time tracking of interventional devices using projection-reconstruction imaging with echo-dephasing (pride). Magnet Reson Med. 2009;62:935–42. Google Scholar
Schaefers G. Testing MR safety and compatibility: an overview of the methods and current standards. IEEE Eng Med Biol Mag. 2008;27:23–7. Google Scholar
Yu N, Gassert R, Riener R. Mutual interferences and design principles for mechatronic devices in magnetic resonance imaging. Int J Comput Assist Radiol Surg. 2011;6:473–88. Google Scholar
Güttler F, Heinrich A, Teichgräber U. Software development for the determination of susceptibility artefacts in MRI after ASTM F2119. Biomed. Technik. 2012;57:480–480. Google Scholar
Chen C, Chen W, Goodman S, Hargreaves BA, Koch KM, Lu W, et al. SEMAC and MAVRIC for artifact-corrected MR imaging around metal in the knee. Proceedings of the ISMRM, Stockholm, Sweden, 130; 2010. Google Scholar
Liao W, Deserno TM, Spitzer K. Evaluation of free non-diagnostic dicom software tools. In Medical Imaging. p. 691903–691903. International Society for Optics and Photonics; 2008. Google Scholar
Dössel O. Biomechanic models and image guided interventions. Biomed Tech (Berl). 2015;60:519. Google Scholar
Heinrich A, Guettler F. Software-based automated measurement of susceptibility artifacts on magnetic resonance images. Google Scholar
Sameni R, Shamsollahi MB, Jutten C, Clifford GD. A nonlinear bayesian filtering framework for ECG denoising. IEEE Trans Biomed Eng. 2007;54:2172–85. Google Scholar
About the article
Published Online: 2016-09-30
Published in Print: 2016-09-01
Citation Information: Current Directions in Biomedical Engineering, Volume 2, Issue 1, Pages 427–431, ISSN (Online) 2364-5504, DOI: https://doi.org/10.1515/cdbme-2016-0095.
©2016 Alfredo Illanes et al., licensee De Gruyter.. This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. BY-NC-ND 4.0